I was wondering how it tricked the game into recognizing the correct answer. Does it end each question by writing the correct answer inside an already darkened portion maybe?
The tracking pays more attention to the length and position of your strokes, rather than doing what you might think of as OCR on the finished characters.
Here's a tool-assised run that the TASBot run was clearly inspired by: https://www.youtube.com/watch?v=_p_UtfOmp9E - you'll notice there's no opportunity for the actual numbers to be drawn in many instances.
For the most part, the images were selected/drawn in a way that fools the engine. The team responsible for the speedrun made a brief writeup that explains some of the stuff they ended up using: http://tasvideos.org/GameResources/DS/BrainAgeTrainYourBrain...
Comments
I was wondering how it tricked the game into recognizing the correct answer. Does it end each question by writing the correct answer inside an already darkened portion maybe?
The tracking pays more attention to the length and position of your strokes, rather than doing what you might think of as OCR on the finished characters.
Here's a tool-assised run that the TASBot run was clearly inspired by: https://www.youtube.com/watch?v=_p_UtfOmp9E - you'll notice there's no opportunity for the actual numbers to be drawn in many instances.
A Japanese language article that goes more into the specifics: http://dic.nicovideo.jp/a/tas%E8%AA%9E
For the most part, the images were selected/drawn in a way that fools the engine. The team responsible for the speedrun made a brief writeup that explains some of the stuff they ended up using: http://tasvideos.org/GameResources/DS/BrainAgeTrainYourBrain...